Anyway, there is no indication that your data would be exposed if you had a private repo that was always private.
I found the article title pretty misleading. There's perhaps an interesting conversation to have around the possibility of wanting to be able to scrub training data after-the-fact and how (or if) that could work - but that's not what the article's headline tried to convey.
It’s like saying Wayback machine is exposing private data because it was captured when it was public.
What an absolute waste of an item on the HN front page.
And what a surprise! So does Bing: https://www.bing.com/webmasters/help/bing-content-removal-to... (As do Google etc.)
Neither of these expunging mechanisms work if you're not the domain owner, however, so this is just one more reminder that any content you upload to somebody else's website is never fully under your control.
Or maybe the post needs a better title